Wavelet-artificial neural network model for water level forecasting

Forecasting water level is used in many domain in our real life such as agriculture, flood forecasting, hydro electrical dam. This problem is very importance in Vietnam agriculture. This paper addresses daily water level forecasting models with short time such as 12, 24, 48, 72 or 168 hours. The methodology of forecasting is combining Wavelet-Artificial neural network (WAANN) that encourages the advantage of each model Wavelet analytic and ANN. The method to combine two model Wavelets and ANN and the hybrid algorithm are described in this paper. In application, the proposal model is applied to forecast water level in Yen Bai station, northwest Vietnam. Input variables into Wavelet-ANN structure are time series of water. Wavelet analysis removes the random noise that called high signal of time series data. The left of time series data is used ANN model to forecast. The results of ANN and WAANN are built then compare their performances. To evaluate forecasting result, the mean square error (MSE) and mean absolute deviation (MAD) are considered. The results of WAANN of water level forecasting shows better performance than ANN.